The document and email API for commercial insurance

Point it at a submission email, an application, a policy, or a binder and get back structured, sourced insurance data at up to 99% accuracy. Build on it instead of building it.

You can build it. Precision is where it gets hard.

Standalone extraction gets you most of the way, then the cost shows up in tokens and engineering time as you chase the accuracy a regulated line of business actually requires. Most builds stall there, often without a reliable measure of how close they got.

Insurance-native extraction as an API

Send an email or a document, get back clean JSON with thousands of normalized fields, each one sourced to the page it came from.

Anzen uses frontier and custom-trained models tuned against five years of insurance-expert data, improving daily as customers transact. A test bed of billions in transacted submissions lets us benchmark new models the day they release and optimize token spend, so you are not paying to find out what works.

The documents you actually transact on

All commercial lines, 36 document types classified on intake.

Submissions & renewals

Read the email thread and every attachment as one submission. Know which file is the application, which is the loss run, which is the prior policy.

Applications

ACORDs, carrier supplementals, and custom forms reconciled to one normalized set, so every answer is comparable across the market.

Policies & binders

Sublimits per insuring agreement, endorsements by form number and version, the exclusions and definitions that decide coverage.

Quotes & loss runs

Structured, queryable, ready to compare.

A data layer, not another UI

Point it wherever the work already happens.

New business

Classify and route submissions the moment they hit the inbox, prefill your AMS or PAS, triage by appetite.

Servicing

Read endorsement requests as they come in and trigger the right action automatically.

Business intelligence

Pipe normalized submission and policy data into your warehouse to see where your book actually sits.

Insurance is not a generalist problem

Most extraction tools are horizontal. They are good generalists, and insurance is not a generalist problem. It is regulated, specific, and the whole game is getting reliably past ninety-five percent.

We are built vertical. Our context graph knows thousands of carriers, wholesalers, and agencies. Our team has run brokerages and built data systems, so we read documents the way an operator does and ship them the way an engineer expects.

99%accuracy on the fields that decide coverage
1,000sof carriers, wholesalers & agencies in the context graph
$1B+of quoted premium battle-tested against
every value sourced to the page it came from

See it on your messiest thread

We'll walk you through the API on your own documents if you'd like. No integration required to see it work.